The Khan List - Apple-Google AI Pact, NVIDIA Vera Rubin, OpenAI's $110B Round

🔵 The Big Story

Anthropic won't budge as Pentagon escalates AI dispute

Anthropic publicly named DeepSeek, Moonshot, and MiniMax for orchestrating 24,000+ fake accounts to systematically extract Claude's capabilities. This is the first major AI company to go on record accusing rivals of industrial-scale model theft.

Why this matters: Distillation attacks let smaller labs skip billions in training costs by extracting a frontier model's reasoning patterns through clever prompting at scale. If proven, it fundamentally changes the economics of AI competition — and could trigger export controls on API access itself, not just chips.

What to watch: Congress will likely hold hearings within weeks. Expect other frontier labs (OpenAI, Google) to release similar findings. The AI IP legal framework is about to be stress-tested.

Read the full story →

Boeing Runs LLMs on Space-Grade Hardware — A First for Orbital AI

AI inference on radiation-hardened satellite processors. Opens the door to autonomous satellite operations and real-time Earth observation without ground station dependency.

Read more →

Pentagon Strikes Deal to Deploy Grok in Classified Systems

First classified deployment for Musk's xAI. Raises questions about vendor concentration in defense AI and the growing overlap between Silicon Valley founders and the national security apparatus.

Read more →

Bloomberg · Market breadth narrows as AI momentum names give back gains

TechCrunch · At least a dozen VCs are hedging their frontier AI bets

CNBC · Sentiment shift from euphoria to anxiety in AI-adjacent equities

NYT · Defense officials push back on Anthropic's usage restrictions

Gizmodo · When a viral AI essay becomes a market-moving event

Today's challenge: Pick one AI tool from today's stories and spend 15 minutes testing it against a real workflow. Document what worked, what didn't, and whether it saves time or money. The best AI insights come from doing, not reading.

📡 Signal: Anthropic's distillation disclosure is backed by server logs, account forensics, and named entities. This isn't speculation — it's documented industrial-scale IP extraction that will reshape API access controls industry-wide.

📢 Noise: "AGI is 6 months away" — This claim resurfaces every quarter from different sources. Current frontier models still fail at basic spatial reasoning, multi-step planning, and novel scientific discovery. Progress is real. The timeline hype is not.

Q&A of the Day

"If Anthropic can detect distillation attacks, why can't they just block them in real time?"

They partially can — and do. Anthropic already rate-limits suspicious patterns and has shut down thousands of accounts. The challenge is that sophisticated distillation looks identical to legitimate high-volume API usage. The attackers rotate accounts, vary prompt patterns, and use distributed infrastructure. It's the same cat-and-mouse game as ad fraud or bot detection: you catch most of it, but the 5% that slips through at scale still extracts meaningful model capability. The real fix isn't technical — it's legal and regulatory, which is why Anthropic went public.

Got a question? Reply to this email — we answer one every issue.

Know someone drowning in AI noise?

Forward this email. 5 minutes of signal beats 5 hours of scrolling.